Portkey

A production-grade platform for generative AI that unifies access to multiple models via an AI gateway, equipped with observability, guardrails, governance and prompt management, and has now been renamed PRISMA AIRS AI Gateway.

AI Developer Tools#LLMOps#Gateway#Observability

Updated: 2026-10-26

What is Portkey?

Portkey is positioned as the "Production Stack for Gen AI Builders". The headline on its official website announces that it has now been renamed PRISMA AIRS AI Gateway and is open to all enterprises. The problem it aims to solve is: after models are integrated into business operations, who will uniformly manage access, stability, costs, security and compliance. The official answer is "Observe, govern, and secure every AI interaction across your enterprise" — to observe, govern and secure every AI interaction within the enterprise, while "democratizing" Gen AI capabilities across the entire organization (Democratize and productionize Gen AI across your entire org), so that teams do not have to connect to models independently.

Who is it for?

Medium and large teams that have already integrated large models into production systems and are beginning to pay attention to costs, stability, and compliance; organizations that need to uniformly manage multiple model vendors; and platform teams that want to quickly promote Gen AI capabilities across the entire company.

Key features

AI Gateway

Unifies the entry and exit points to manage multiple model vendors. The official solution specifically proposes the "AI Gateway Pattern" practice: a single interface connects all models, and switching vendors does not require rewriting code.

Observability

Observe and record every AI interaction. The official states it aims to be the "single source of truth for LLM pricing," which is relied on for troubleshooting and cost attribution.

Guardrails and Governance

Guardrails impose security and compliance constraints on inputs and outputs, AI Governance provides governance capabilities for model usage and access, and the official emphasizes end-to-end LLM Orchestration.

Prompt Management

Centralize prompt version management, decouple from code to prevent prompts from being scattered and untraceable; combined with gateway logs, you can also compare the effects of different prompt versions under real traffic.

Minute-level Integration & MCP Security

The official emphasizes "Integrate in a minute" for integration within minutes; meanwhile, it provides security control over access to MCP tools to reduce risks associated with tool invocation.

Enterprise Recognition

The official disclosed that it is adopted by Fortune 500 companies and startups, and has gained recognition from industry leaders.

How to use Portkey

First, integrate all calls to existing models into the gateway to create a unified entry point and establish the baseline for costs and calls.
Then, gradually enable capabilities such as guardrails, governance, and prompt management.
For guardrails, it is recommended to start with configuration in sensitive business scenarios.
It is advisable to begin with a pilot in one business line.
after verifying stability and team collaboration methods, roll it out to the entire organization.
Prepare a backup channel for the gateway before launch, and configure retry and downgrade strategies for calls to key models to prevent full-link failures caused by gateway jitter.

Things to know

The brand and product name has been changed (Portkey → PRISMA AIRS AI Gateway), and it is recommended to use both names when searching for relevant information.

The gateway is the only path for all traffic. When selecting a gateway, focus on its stability and failover capabilities, as a single point of failure can have a significant impact.

In the cloud service mode, confirm data residency and compliance clauses for calls involving sensitive data.

Full capabilities such as governance and MCP security are biased towards the enterprise edition; if individuals or small teams only need the gateway, pay attention to the pricing tiers.

The platform has multiple modules. It is recommended to start with the gateway as a single point of entry and gradually enable other modules, which is easier to implement than full access at once; individual developers can also benefit by using it solely as a multi-model access layer.

If you only need model routing and logging, first confirm whether the free tier and trial quota cover your actual usage to avoid unexpected increases in monthly bills.